I vaguely remember about some benchmarking project that deliberately randomised these compiler decisions, so that they could give you more stable estimates of how well your code actually performed, and not just how well you won or lost the linker lottery.
There was Stabilizer [1] which did this, although it is no longer maintained and doesn't work with modern versions of LLVM. I think there is something more current now that automates this, but can't remember what it's called.
As already mentioned this is likely Emery Berger’s project with the idea of intentionally slowing down different parts of the program, also to find out which parts are most sensitive to slowdowns (aka have the biggest effect on overall performance), with the assumption that these are also the parts that profit the most from optimisations.
